Suicidal thoughts in preadolescents linked to altered brain reward responses

In one of the largest brain-imaging investigations of suicidal thinking in children conducted to date, scientists report that suicidal thoughts in nine- and ten-year-old children leave a measurable imprint on the brain’s reward circuitry — and that the imprint looks radically different depending on who reports the thoughts. Analyzing functional MRI scans from 5,702 preadolescents in the Adolescent Brain Cognitive Development (ABCD) Study, researchers found that children whose suicidal ideation was endorsed by both themselves and their parents showed blunted activation across broad frontal regions while anticipating monetary rewards and losses. Children flagged by only one reporter — the child alone or the parent alone — showed the opposite pattern: heightened activation in those very regions during anticipation. The study, published open access in Child Psychiatry & Human Development, argues that suicidal ideation in early preadolescence is not a single brain state but several distinct ones, and that the identity of the informant may be the clue that separates them.

The urgency behind the work is stark. Suicide is the second leading cause of death among United States children and young adolescents aged 10 to 14, and preadolescent suicide increased by an average of 8.3 percent annually between 2008 and 2022. Mortality then climbs steeply with age, from 2.8 per 100,000 people at ages 10 to 14 to 10.57 per 100,000 at ages 15 to 19. Suicidal ideation is one of the strongest known predictors of a subsequent suicide attempt, and an estimated 17 percent of preadolescents who report such thoughts will go on to attempt one. Yet almost nothing is known about the neural mechanisms of suicidal thinking at ages 9 and 10 — the developmental window that opens just before the surge in attempts seen in later preadolescence and adolescence. Understanding what happens inside the maturing brain at this precise moment, the authors argue, is essential for building interventions before risk escalates.

The study homes in on reward processing, one of the most intensively studied systems in psychiatric neuroscience. Decades of work have tied suicide risk to disruptions in how the brain computes value, assigns salience and guides decisions. Across monetary and risk-based paradigms, adults with a history of suicidal thoughts or behaviors have exhibited both increased and decreased recruitment of reward-related regions — the subgenual anterior cingulate cortex, the insula, the orbitofrontal cortex and the caudate — with the direction of effect shifting by population and task condition. Studies in adolescents tell an equally tangled story: suicidal ideation has been linked to blunted activation during high-risk conditions, reduced striatal response to predictable gains, heightened salience responses under uncertainty, and exaggerated reactions to negative social feedback and self-appraisals. One suspect behind the inconsistency, the researchers reasoned, is methodological: most studies collapse suicidal ideation into a single group, and nearly none had probed the brains of the youngest children.

Indeed, only one prior study had directly examined reward-related brain activation in preadolescent suicidality. Using the ABCD baseline sample, Vidal-Ribas and colleagues in 2021 tested 167 predefined regions of interest during a reward task and found no significant associations with youth- or parent-reported lifetime suicidal thoughts and behaviors; the sole corrected finding was thinner cortex in the left superior temporal sulcus for parent-reported suicidality, a region far removed from reward circuitry. The new study, led by Alyssa J. Parker, Annika J. Quam, Johanna C. Walker, Jillian Lee Wiggins and Lea R. Dougherty of the University of Maryland and San Diego State University, departed from that constrained approach in two ways. First, it replaced predefined regions with flexible, data-driven whole-brain analyses capable of surveying the entire brain. Second, it treated reporter identity as a variable in its own right. Parents and children agree remarkably rarely about whether a child has experienced suicidal thoughts; in this sample, agreement between the two reporters was low (kappa = 0.14), suggesting each informant may capture a different behavioral reality.

The data came from the baseline wave of the ABCD Study, a longitudinal project following more than 10,000 American children from age 9 or 10 into early adulthood. From 11,874 children in data Release 4.0, 5,702 participants with a mean age of 9.96 years survived stringent exclusions for incomplete records, poor task performance, failed fMRI quality checks and histories of brain trauma or seizures. Each child performed the Monetary Incentive Delay task inside a 3-tesla MRI scanner at one of 21 sites. In this paradigm, an anticipatory cue lasting two to six seconds signals whether a fast response to an upcoming target will win money, lose money or carry no monetary stakes; a feedback screen lasting 1.5 to 1.8 seconds then reveals whether the child hit or missed the target. Difficulty adapts dynamically so every child succeeds on roughly 60 percent of trials, standardizing the challenge across the sample. Preprocessing included motion correction, denoising, band-pass filtering, censoring of high-motion time points and spatial smoothing with a 6-millimeter kernel, with anticipation and feedback modeled using the dmBLOCK approach. The team then used AFNI’s 3dMVM software to test how reporter group, task condition and performance interacted to shape activation across the whole brain, entering scanner serial number as a covariate to absorb hardware differences between sites. Significant clusters required a voxel-wise threshold of p = .001 and a minimum of 37 voxels, corrected for multiple comparisons through family-wise error rate. Follow-up models in R, nested by scanner and family and controlling for sex at birth, pubertal development, age and parental education, applied robust mixed-effects statistics with false-discovery-rate-corrected comparisons. The sample divided into four groups: 5,294 dyads in which neither reporter endorsed ideation, 200 in which only the child did, 169 in which only the parent did, and 39 in which both did.

During the anticipation period, whole-brain analysis uncovered significant group-by-condition interactions in seven clusters spanning the left middle frontal, superior medial and superior parietal gyri, the right angular, temporal and postcentral gyri, and the bilateral lingual gyrus. The 39 children endorsed by both reporters stood out sharply: relative to every other group, their activation was significantly blunted while anticipating both rewards and losses across these frontal, parietal and occipital regions. The picture inverted for the discordant groups. Children flagged by a single informant showed heightened anticipation activation compared with children free of reported ideation, and the parent-only group exceeded even the child-only group in several regions during loss anticipation. Main effects of reporter group also surfaced in the left precuneus and middle temporal gyrus, where children with reported ideation generally out-activated the no-ideation group, with the parent-only group often highest of all. The effect sizes were statistically modest — partial eta-squared values between 0.002 and 0.004 — but they emerged consistently in a sample of nearly six thousand and withstood covariate adjustment.

The feedback period told a different and, in some respects, more surprising story. Fifteen clusters spanning frontal, temporal, cingulate, parietal, occipital and subcortical territory showed group-by-condition interactions once outcomes were revealed. Here the both-reporter group reversed direction: after receiving rewards and losses, its activation rose above that of the no-ideation group and both single-reporter groups, spanning the right middle and superior temporal gyri, the inferior occipital and fusiform gyri, and the left cingulate, middle cingulate, insula, thalamus, supramarginal gyrus and rolandic operculum. Yet during reward receipt specifically, the same children showed significantly reduced activation in the left superior parietal lobe relative to all other groups, and the no-ideation group generated stronger left insula responses than the child-only group when receiving losses. The insula, thalamus and prefrontal regions implicated here, the authors note, mark candidate targets for neurologically informed interventions.

To the authors, these two patterns describe distinct neurophenotypes of early suicidal ideation. When both a child and a parent acknowledge the thoughts — typically the more severe or outwardly visible presentations — the brain under-responds while anticipating incentives, echoing prior reports of blunted reward responsiveness in youth with suicidal ideation, from EEG recordings to fMRI studies of predictable gains and positive self-appraisal. Blunted reward reactivity is also linked more broadly to psychopathology, childhood adversity and vulnerability to addiction. When only one informant endorses, however, the brain over-responds, both while anticipating outcomes and after receiving them — a profile reminiscent of heightened reward salience, a mechanism implicated in impulsivity, substance use and risk-taking. This reporter-dependence, the team argues, may explain why decades of research on reward processing and suicidality have pointed in contradictory directions: pooling informants fuses biologically distinct groups into a single average that resembles neither. The conclusion dovetails with earlier ABCD analyses in which parent-reported, but not youth-reported, suicidality predicted cortical thinning, while both reporters were tied to altered default mode network connectivity.

The study carries important caveats. The both-reporter group contained only 39 children, against 5,294 with no reported ideation — an imbalance that constrains statistical power and generalizability. Suicidal ideation was captured with dichotomous items from a structured diagnostic interview rather than a continuous severity scale, and excluded participants were slightly more likely than included ones to have child-reported ideation. Because the design is cross-sectional, it cannot establish which informant better forecasts future risk, or how these neural signatures evolve across adolescence. Supplementary analyses controlling for concurrent internalizing and externalizing symptoms, and substituting passive for active ideation, left the pattern largely intact.

The clinical implications are already taking shape. For children whose ideation is recognized by both themselves and their parents, the blunted anticipation response points toward interventions that help youth attend to, savor and value rewarding experiences — approaches aimed at the anhedonia-like profile long associated with dampened reward function. For single-reporter presentations, the exaggerated responses suggest the opposite need: helping children regulate an amplified salience response to incentives and losses, the same circuitry implicated in impulsivity and early substance use. Above all, the authors contend, assessment must be multi-informant. Parent report alone or child report alone may each miss a distinct subgroup, whereas the two reports together sketch a fuller neural map of early suicidal thinking. Supported by a National Institutes of Health grant, the team now calls for longitudinal studies tracing how these reward signatures unfold across development — work aimed at catching, and redirecting, the brain states that precede adolescent suicide attempts.